A report by ABI says such devices will feature deep learning models to automate and augment decision-making in applications such as intelligent traffic management.
Security cameras with artificial intelligence (AI) will be the norm for smart city applications such as intelligent traffic management and preventative threat detection, according to latest research.
The global tech market advisory firm, ABI Research’s analysis report, Deep Learning-Based Machine Vision in Smart Cities, reveals that the installed base of smart cameras with an AI chipset will reach over 350 million in 2025. More than 65 per cent of cameras shipped in 2025 are expected to come with at least one AI chipset.
Such cameras will feature deep learning (DL) models to automate and augment decision-making in applications such as intelligent traffic management, autonomous asset, pedestrian flow monitoring and management, physical and perimeter security, and preventive threat detection.
“More and more city and county governments around the world are actively looking to utilise AI. This has led to a boom in the adoption of smart cameras with edge AI chipsets,” said Lian Jye Su, principal analyst of AI and machine learning, ABI Research.
Aside from low latency, data privacy concerns have also driven the adoption of AI at the edge as information can be processed without being sent to the cloud. The report lists Ambarella, HiSilicon, Intel, NVIDIA, Qualcomm and Xilinx as some of the key AI chipset suppliers in the smart city space. Increasingly, TinyML vendors that offer always-on machine vision are expected to play a key role in enabling always-on machine vision through battery-operated cameras, Lidar, infrared, and other sensors.
According to ABI, most of these workloads are performed by either DL models hosted in the cloud, offered by video analytics vendors such as SenseTime, Ipsotek, icentana, and Sentry AI, or DL inference in smart cameras and network video recorders, such as HikVision and Dahua. Both deployment methods come with their own respective strengths and weaknesses.
“More and more city and county governments around the world are actively looking to utilise AI. This has led to a boom in the adoption of smart cameras with edge AI chipsets”
Su explained that two technology trends will likely further catalyse the deployment of DL-based machine vision: “The first one is edge computing. Instead of deploying specific DL models on smart cameras that are multiple times more expensive than legacy cameras, city and county governments can host DL models on gateways and on-premise servers. This allows data to be processed and stored at the edge, providing faster response time than relying on cloud infrastructure.
“The second is 5G. While network slicing won’t be commercially ready by 2023, the network slicing capability of 5G allows communication service providers to offer dedicated network resources to host microservices, six nines reliability service assurance, seamless device connectivity, and onboarding to support DL-based machine vision in the smart city.”
The report highlights though that public trust and regulations related to adopting AI in public cameras is a big challenge facing their implementation. The public and human rights advocates around the world are wary of misuse and have been pushing back against the adoption of facial recognition technologies.
“Trust is a critical component in public safety technologies. ABI Research encourages developers, vendors, authorities, and the general public to focus on constant dialogue and the introduction of common technology platform for transparency, as well as AI ethics and governance frameworks that can minimise biases,” added Su.
He continued: “Moving forward, the technology vendors that are successful in the smart city are those which will be able to demonstrate transparent and explainable DL models and those who show a willingness to embrace open and common standards and ethical frameworks.”
You might also like: